On 03/02/2014 07:37 PM, Michał Górny wrote:
> Hi,
> 
> Few months ago I have written a small FAQ on how to use slots
> and subslots for library dependencies properly [1]. However, today
> I see that most of the developers didn't care to properly update their
> packages and when I introduced binary compatibility slot in libgcrypt,
> I had my hands full of work fixing the mess for a single package.
> 
> Honestly, it's PITA to update and commit a few dozen ebuilds just to
> modify a single dependency. Plus sometimes portage's dynamic-deps
> no longer work so I'd have to revbump all the packages as well to
> be 100% correct. And the sole fact that I'm fixing just one dep when
> there's a dozen libraries more that may need fixing in the future...
Please carefully consider whether a revbump is unconditionally required
for every situation.

Pinning a dependency to SLOT 0 will be picked up by dynamic-deps, and if
the user has it turned off it is likely that more building will be
caused by the rebuilds versus the new binary compatibility slot.

dynamic-deps does not apply to the subslot dependency operator, but
revbumping to add it will just cause the user to have to build the
package twice - once for the revbump and once when the subslot changes.

> 
> So, I'm asking: would you mind if I started taking random packages
> and updating the library dependencies (whenever they are clear) to use
> slot :0 (in EAPI 1..4) and :0= (in EAPI 5) as appropriate?
> 
> [1]:http://article.gmane.org/gmane.linux.gentoo.devel/88541
> 

For adding an explicit slot after a binary compatibility slot is
introduced, I think that is fine. I have been doing that for
virtual/jpeg as I come across them (AFAIK no effort was made to fix this
when it was introduced) and I have never had any complaints.


Reply via email to